ABSTRACT
Inthesepapers,asastandard,40replicaswereusedtocreatethenull distribution.Thedistributionwassmoothedbyanonparametricintegrated kerneltypeestimateandthecriticalregioncorrespondingtoana-level testisthenobtainedbyselectingthe(I-a)-quantilec~ofthebootstrap
Ideally one would like to analyse the randomization/bootstrap approach theoretically. Asymptotic theory is lacking for the bootstrap in this situation, but work is now in progress, Neumann and Kreiss (1996). It is a highly nontrivial task, and large sample sizes may again be needed to check the validity of the asymptotic theory.
